Last Thursday I gave a talk on AI governance at the SMU School of Law, including how countries converge and diverge on data governance issues in AI – thank you A/Prof Han-Wei L. for the invite! 3 thoughts to share: 1) “Interoperability” – many guidelines and frameworks from countries (e.g. Singapore, EU, Canada, US) and international organisations (e.g. OECD) reference this term but what does it mean? I’ve yet to find a definition/state where AI governance frameworks are said to be interoperable, but bearing in mind that the goal is "harmonisation of international AI governance frameworks to reduce industry’s cost to meet multiple requirements" (per SG’s IMDA and the USA’s NIST in the joint mapping exercise between Singapore's AI Verify and USA's AI Risk Management Framework) – perhaps interoperability will be achieved only at a granular level in the form of what steps must be taken and how to measure whether those steps have actually been taken, than at a high-level principles approach (e.g. fairness, explainability, transparency, safety). I’m excited that Singapore has just mapped AI Verify with ISO/IEC 42001:2023 as well! Perhaps 'interoperability' will be through international standards (as they may be more readily adopted by countries than the standards of another country). 2) At the same time, standards and testing metrics must also take into account the unique socio-economic climate of each country. Singapore’s ‘Cataloguing LLM Evaluations Paper’ notes that the present framing of toxicity, bias and demographic considerations in LLM evaluations tends to be Western-centric. The KoBBQ paper (May 2024) also notes that the effectiveness of tests to assess social biases of LLMs depends on the cultural context – e.g. drug-use is associated with low SES in some countries, but with high SES in Korea. 3) An interesting divergence is in copyright over AI-generated output – the US Copyright Review Board in Jason Allen’s case said ‘no’ because having to prompt 624 times before receiving a satisfactory image shows how ‘random’ the output is such that the style and other elements of authorship are determined by the AI system and not the human. In contrast, the Beijing Internet Court in Li v Liu (where the plaintiff had ~20 prompts for what he wanted (‘Japan idol’, ‘reddish-brown plaits hairs’) and ~120 prompts for what he did not want (‘closed eyes’)) found that this reflected his “aesthetic choice and personal judgment” and copyright subsisted. Our IPOS x SMU Landscape Report characterizes these decisions as the debate evolving beyond a fact-centric question of how much human input and control is needed to “a question of the fundamental concept of originality to attribute human ownership”. I agree, but personally think the outcome could also be driven by pragmatism – in the Allen case the plaintiff ‘won a prize’ and merely sought to register his copyright, but in Li the plaintiff ‘was plagiarized’ so a remedy would be necessary.
Last week, we had an engaging talk by Cheryl Seah, a leading practitioner in data governance and law and tech from Drew & Napier LLC, on ‘𝐀𝐈 𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞: 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥𝐢𝐭𝐲 𝐚𝐧𝐝 𝐈𝐧𝐭𝐞𝐫𝐨𝐩𝐞𝐫𝐚𝐛𝐢𝐥𝐢𝐭𝐲’. Cheryl delved into the ASEAN approach to AI and how it measures up to global AI governance standards, with a spotlight on new developments in AI copyright issues. The session was insightfully moderated by CAIDG's Associate Prof. Han-Wei L.. We thank Josh Lee Kok Thong, CAIDG Research Affiliate and Future of Privacy Forum’s Managing Director of APAC, for sharing his hands-on experience in AI governance from a policymaking perspective earlier this semester, providing valuable insights amid the changing geopolitical landscape. #DataGovernance #AIGovernance #Internationality #Interoperability
Han-Wei L.
3mo
Thanks very much Cheryl Seah!
Associate Professor of Law and Director, Centre for AI & Data Governance
3moThanks Cheryl for joining us, and sorry I missed your talk! Btw just to clarify because it's a SMU Centre for AI & Data Governance post--the Jason Allen in the case mentioned above is not me! (There is also a Jason Allen at the Bank of Canada, and a I think a NFL player!).